involvement of spearman's g in conceptualisation versus...

Acta Psychologica 170 (2016) 112–126

Contents lists available at ScienceDirect

Acta Psychologica

j ourna l homepage: www.e lsev ie r .com/ locate /actpsy

Involvement of Spearman's g in conceptualisation versus execution ofcomplex tasks

Ellen L. Carroll 1, Peter Bright ⁎Department of Psychology, Faculty of Science & Technology, Anglia Ruskin University, East Road, Cambridge CB1 1PT, UK

⁎ Corresponding author.E-mail address: [email protected] (P. Bright).

1 Present address: Division of Anaesthesia, School of CCambridge, Box 93, Addenbrooke's Hospital, Hills Road, C

http://dx.doi.org/10.1016/j.actpsy.2016.06.0110001-6918/© 2016 The Authors. Published by Elsevier B.V

a b s t r a c t
a r t i c l e i n f o
Article history:Received 1 September 2015Received in revised form 24 June 2016Accepted 27 June 2016Available online xxxx

Strong correlations between measures of fluid intelligence (or Spearman's g) and working memory are widelyreported in the literature, but there is considerable controversy concerning the nature of underlyingmechanismsdriving this relationship. In the four experiments presented here we consider the role of response conflict andtask complexity in the context of real-time task execution demands (Experiments 1–3) and also address recentevidence that g confers an advantage at the level of task conceptualisation rather than (or in addition to) task ex-ecution (Experiment 4). We observed increased sensitivity of measured fluid intelligence to task performance inthe presence (vs. the absence) of response conflict, and this relationship remainedwhen task complexity was re-duced. Performance-g correlations were also observed in the absence of response conflict, but only in the contextof high task complexity. Further, we present evidence that differences in conceptualisation or ‘modelling’ of taskinstructions prior to execution had an important mediating effect on observed correlations, but only when thetask encompassed a strong element of response inhibition. Our results suggest that individual differences in abil-ity reflect, in large part, variability in the efficiency with which the relational complexity of task constraints areheld in mind. It follows that fluid intelligence may support successful task execution through the constructionof effective action plans via optimal allocation of limited resources.

© 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license(http://creativecommons.org/licenses/by-nc-nd/4.0/).

Keywords:Spearman's gFluid intelligenceResponse inhibitionTask complexityTask modellingChunking

1. Introduction

Strong correlations between performance on tests of workingmem-ory capacity (WMC) and fluid intelligence (g) are well established(e.g., Ackerman, Beier, & Boyle, 2002; Conway, Cowan, Bunting,Therriault, & Minkoff, 2002; Süß, Oberauer, Wittmann, Wilhelm, &Schulze, 2002; Unsworth, Redick, Lakey, & Young, 2010). Themediatingfactors in this relationship, however, are not fully understood. Tradition-ally, theworkingmemory (WM) systemhas beenpresented as amentalworkspace associated with the concurrent storage and processing of in-formation; Baddeley and Hitch's (1974) multicomponent WM model,for example, comprises domain-specific storage buffers and a centralexecutive. Complex span tests, which typically assess memory forwords or digits in the face of a demanding interleaved task are amongthe best measures of WMC and are also sensitive to variations in fluidintelligence. In contrast, simple span tests, which do not encompass ad-ditional processing demands, are typically weakly correlated withmea-sures of WMC and intelligence (e.g., Conway, Kane, & Engle, 2003;Daneman & Carpenter, 1980; Engle, Tuholski, Laughlin, & Conway,

linical Medicine, University ofambridge CB2 0QQ, UK.

. This is an open access article under

1999; Turner & Engle, 1989). This finding has led some authors toargue for the central role of processing (e.g., executive attention;Conway et al., 2003; Engle et al., 1999) in driving the correlation be-tweenWMC and g. Subsequent evidence, however, supports the centralrole of short-term storage (e.g., immediate recall of memory for num-bers, letters, or visual arrays; Chuderski, Taraday, Nęcka, & Smoleń,2012; Colom, Abad, Quiroga, Shih & Flores-Mendoza, 2008; Colom,Flores-Mendoza, Quiroga, & Privado, 2005).

The executive attention account of inter-individual differences inWMC (e.g., Engle & Kane, 2004; Kane, Conway, Hambrick, & Engle,2007) claims that individuals with lowWMC have relatively limited ca-pacity for controlling goal-directed attention, and for resolving responseconflict, compared to individuals with high WMC. According to thisview, high WMC individuals typically produce fewer errors on taskssuch as the classic Stroop (1935) test (e.g., Kane & Engle, 2003; Long &Prat, 2002) because they possess relatively greater capacity for directingattention to naming the colour, and for resolving the competition be-tween eliciting the prepotent (but incorrect) response of reading theword and producing the appropriate response of naming the colour inwhich the word is written. On the basis of this executive attention ac-count the memory maintenance and retrieval theory of WMC hasbeen developed (e.g., Unsworth & Engle, 2007a, 2007b; Unsworth &Spillers, 2010) which claims that high WMC individuals are better atboth maintaining relevant information in primary (working) memory,

the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

http://crossmark.crossref.org/dialog/?doi=10.1016/j.actpsy.2016.06.011&domain=pdf

0opyright_ulicense

http://dx.doi.org/10.1016/j.actpsy.2016.06.011

mailto:[email protected]

Journal logo

http://dx.doi.org/10.1016/j.actpsy.2016.06.011

0opyright_ulicense

http://www.sciencedirect.com/science/journal/00016918

www.elsevier.com/locate/actpsy

113E.L. Carroll, P. Bright / Acta Psychologica 170 (2016) 112–126

and at using appropriate retrieval cues to retrieve information from sec-ondary (long-term) memory when required. Other researchers stressthat response inhibition is the singlemost important factor in individualdifferences inWMC (Hasher & Zacks, 1988; Lustig, May &Hasher, 2001;May, Hasher, & Kane, 1999). The claim is that individuals with highWMC are better at restricting WM access to task-relevant information,resolving response conflict, and inhibiting dominant but inappropriateresponses. High WMC individuals therefore perform better on theStroop because they are better able to limit WM access to the relevanttask component (naming the colour) and at inhibiting the dominantbut inappropriate tendency to name the word. In the context of themaintenance and retrieval view, this theory would also explain howconflict is resolved between the inappropriate stimulus-response map-ping of reading the word held in secondary memory and the relevantbut less prepotent mapping of stating the colour held in primary mem-ory. The three accounts are not mutually exclusive and share overlap-ping theoretical claims, but they also incorporate distinct and testablepredictions (Redick, Calvo, Gay, & Engle, 2011) as outlined below.

Performance on the Stroop is usually considered to reflect capacityfor response inhibition and significant correlations between perfor-mance on the Stroop and psychometric intelligence have been reported(e.g., Dempster, Corkill & Jacobi, 1995; Polderman et al., 2009; Salthouse,Atkinson, & Berish, 2003). Functional magnetic resonance imaging(fMRI) studies indicate that the anterior cingulate is recruited in condi-tions of response conflict (Kerns et al., 2004), and by performance ontasks with high g-loadings such as Raven's Advanced Progressive Matri-ces (Gray, Chabris & Braver, 2003). Cognitive models of “general ability”(e.g., Das, 2002) and prefrontal cortex (Roberts & Pennington, 1996)also highlight the importance of inhibition in intelligence. Nevertheless,studies based on factor analysis have produced inconsistent findings.For example, Salthouse et al. (2003) report a strong relationship (r =0.73) between their composite measures of inhibition and fluid intelli-gence in a large sample of adults (N = 261). Conversely, on the basisof evidence suggesting that executive functions—specifically, inhibitingprepotent responses, shifting between tasks/mental sets, and updatingthe contents of WM—are correlated but separable (e.g., Miyake et al.,2000; Miyake & Friedman, 2012), Friedman et al. (2006) observed thattheir composite measures of inhibition (r = −0.11) and shifting(r=−0.08) did not load significantly onto their fluid intelligence con-struct whereas WM updating did (r = 0.74; WM updating was opera-tionalized by tasks that required the adding and deleting ofinformation held in WM: keep-track, letter-memory, spatial 2-back).Other studies have demonstrated that WMC is not related to the abilityto resist interference or dual-task coordination (e.g., Oberauer, Lange &Engle, 2004). These findings indicate that the correlation between re-sponse inhibition and intelligence is not straightforward, and thereforethat interaction with some other task component(s) may be of criticalimportance in driving the relationship.

Redick et al. (2011) compared the executive attention, maintenanceand retrieval, and inhibition theories ofWMC in the context of go/no-gotask performance. The authors compared two go/no-go tasks – a simpletask involving a “go” response to one letter and a “no-go” response to allother letters (with a reversemapping in another block), and a condition-al task involving a go response to two letters conditional on the currenttarget being different to the last. Differences between individuals withhigh/low WMC were observed only in the conditional task, such thathighWMC individuals performed better on both target trials (target let-ters meeting the conditions for a go response) and lure trails (target let-tersmeeting the conditions for a no-go response). Further, performancewas significantly correlated with WMC in the conditional task only.These findings were interpreted in the context of the maintenanceand retrieval account of WMC, with only the conditional task requiringactive monitoring and updating of stimulus-response mappings, andthe retrieval of the appropriate goal-relevant response. Redick et al.(2011) argue that if inhibition or executive attentionwere fundamentalaspects of WMC, differences between individuals with high and low

spans would also be observed in the simple task, because a prepotentresponse must be inhibited or response conflict resolved in both tasks.

An alternative view is that if attention is given to maintaining andupdating the stimulus-response mappings, reduced attention is avail-able for resolving the conflict associated with the no-go requirement,producing more error on these trials. Consistent with the notion ofshared but limited resource availability for processing and storage re-quirements, research has shown that anti-saccade (Mitchell, Macrae, &Gilchrist, 2002; Roberts, Hager, & Heron, 1994) and motor response in-hibition (Hester & Garavan, 2005) capacities decline with increasingWM load. Studies directly addressing storage versus processing ac-counts of the driving force in inter-individual differences in the WMC–g relationship emphasise the overarching importance of storage. For ex-ample, Colom et al. (2008) claim that simple short-term storage(i.e., memory for numbers/letters/visual arrays) accounts for a largeproportion of the relationship between WMC and g, and that althoughattention control, WM updating, and mental speed are independentlycorrelated with g, these relationships disappear when short-term stor-age is controlled for. In a related study, Chuderski et al. (2012) foundthat their storage latent factor (comprising memory for visual arrays,monitoring of relations among stimuli, and updating information inWM) accounted for 70% of the variance in measured fluid intelligence.For their three processingmeasures, only attention control, and neitherresistance to interference nor response inhibition, was correlated withfluid intelligence (accounting for 25% of the variance in fluid intelli-gence), yet, when storage was controlled for, this correlationdisappeared.

A visual change detection study (Fukuda, Vogel, Mayr, & Awh, 2010)has claimed that the number of representations that can be held inWMis highly correlated with fluid intelligence but that the resolution withwhich stimulus representations are stored in WM is largely indepen-dent. Nevertheless, observations by Duncan, Emslie, Williams,Johnson, & Freer (1996) indicate that the relationship between fluid in-telligence and WM cannot be explained on the basis of a straightfor-ward storage function. In their letter monitoring task, participantswere able to recall all task requirements after task completion, but thesensitivity to fluid intelligencewas explained by differences in the capa-bility for responding appropriately to those requirements. Failure toproduce the appropriate response (referred to as “goal neglect”) waslargely restricted to participants scoring N 1 SD below the samplemean on the Culture Fair test of fluid intelligence (Cattell, 1971;Cattell & Cattell, 1973).

Duncan et al. (2008) presented evidence that the efficiency withwhich a task is cognitively modelled or held in mind may be of morefundamental importance to the involvement of Spearman's g than thereal time processing demands associated with the task. Across a seriesof computer based experiments, incorporating a variation of the taskpresented here (Bright, 1998), the authors showed that the form inwhich instructions were presented was the primary factor predictingboth the level of goal neglect and the size of correlation between goalneglect and Spearman's g. Thus, although increased task complexitydid not increase the level of neglect of task demands, an additional“dummy” requirement which had no impact on what participantswere required to do during actual task execution increased level of ne-glect and the strength of the performance–g correlation. This pattern ofresults has been replicated in children using a slightly simplified versionof this feature match task (Roberts & Anderson, 2014). On the basis oftheir findings Duncan et al. (2008) claim that the ability to attend to acomplex “task model – a working memory description of the relevantfacts, rules, and requirements used to control current behaviour”(p. 140) is fundamental to individual differences in g (see alsoBhandari & Duncan, 2014; Dumontheil, Thompson & Duncan, 2010;Duncan, Schramm, Thompson, Dumontheil, 2012).

In the present study we explore the relationship between partici-pants' effectivemodelling of task demands and Spearman's g in the con-text of other candidate “risk factors” for the recruitment of g. In

Fig. 1. Representation of a typical RI Present trial in the colour-shape match task. Black,mid grey, and light grey objects are presented in blue, red, and green, respectively. (Forinterpretation of the references to colour in this figure legend, the reader is referred tothe web version of this article.)

Table 1Definitions of performance measures.

Performancemeasure

Definition

Critical error (RIPresent)

Response to double-matching (NO-GO) pair

Critical error (RIAbsent)

Missed ‘double’ (key press and verbal) response todouble-matching pair

Miss No response to single-matching (GO) pairHand error Response to incorrect side of single-matching pairFalse positivek Response (key press) to non-matching (NEUTRAL) pairFalse positivev (RIAbsent)

Response (verbal) to single- or non-matching pair

Criterion fail Miss or hand error to first two single-matching pairsOR Miss to final single-matching pairOR Three or more false positivesOR Critical error

Reaction time Time taken (ms) to respond to single-matching pair

Note. All motor responses weremadewith the index andmiddle fingers of the right hand.

114 E.L. Carroll, P. Bright / Acta Psychologica 170 (2016) 112–126

Experiment 1 we investigate the relationship between response inhibi-tion and g by comparing performance–g correlations (as well as perfor-mance scores) between a task that requires the inhibition of a prepotentresponse (termed RI Present) and a task that requires an alternative,rather than an inhibited, response (termed RI Absent) in older partici-pants. In Experiment 2 we determine whether performance on the RIPresent task replicates in a younger group of participants. In Experiment3 we examine the importance of task complexity (vs. response inhibi-tion) in performance and performance–g correlations by manipulatingthe number of task demands in RI Present and RI Absent conditions. InExperiment 4 we explore the importance of task modelling in g by di-rectly comparing performance–g correlations on the original RI Presenttask across blocks of trials inwhich frames requiring response inhibitionare either present or absent. In this final experimentwe also explore thesensitivity, to task performance and g, of the form inwhich task rules arepresented to participants; specifically, the same task information is pre-sented either as a set of two rules or a set of four rules in order to clarifythe relationship between WM load, task performance and g within thecontext of a task in which the real time execution demands are heldconstant.

2. Experiment 1: response inhibition versus no response inhibition

In Experiment 1 we examined risk factors for the recruitment of g intask performance. We kept the complexity of task instructions constantand focused on real-time processing demands. Pairs of coloured shapesmatching on zero (neither colour nor shape), one (colour or shape), ortwo (colour and shape) dimensions were rapidly presented on a com-puter screen (see Fig. 1). One group of participants—the RI Presentgroup—were required tomake one of twokey press responses (a “go” re-sponse) to single-matching pairs on the basis of whether the left or theright shape contained a tick, but to suppress a response (“no-go”) todouble-matching pairs (and to also ignore neutral non-matchingpairs). Single-matching pairs were presented with higher frequency(30%) than double-matching pairs (7.5%). Therefore, a go response wasthe dominant response to items matching in colour and items matchingin shape; itemsmatching in both colour and shape encompass two activeresponse tendencies which must be inhibited for correct performance.Another group of participants—the RI Absent group—completed an al-most identical task which differed only in the type of response made todouble-matching pairs. The response inhibition demand was removedfor this group; instead, participants were required to press the relevantkey (left or right) and produce a verbal response. The complexity ofthis task—participants need to decide if items match, on how many di-mensions, and on which side a response should be made—is arguablygreater than the stop-signal (Friedman et al., 2006; Salthouse et al.,2003) and go/no-go (Redick et al., 2011, simple version) tasks thathave previously been reported to show no link between fluid intelli-gence and behavioural or motoric type inhibition. We predicted, albeittentatively, that the task that contained response inhibition require-ments, and the specific task items requiring response inhibition (criticalerrors, see Table 1), would be more sensitive to variations in fluidintelligence.

2.1. Method

2.1.1. ParticipantsSixty-nine neurologically healthy adults, aged between 57 and

72 years, were recruited from the MRC Cognition and Brain SciencesUnit panel of volunteers. The RI Present group comprised 38 adults(15 males, 23 females) aged between 57 and 71 years (M = 63.8,SD = 4.6). The RI Absent group comprised 31 adults (10 males, 21 fe-males) aged between 58 and 72 years (M = 65.1, SD = 4.8). CultureFair error (max = 46) was equivalent across the RI Present (M =17.71, SD = 6.35) and RI Absent (M = 16.74, SD = 6.31) groups,t(67) = 0.63, p = 0.529, d = 0.153. All participants gave informed

consent and were paid to participate. Our primary reason for samplingfrom an older group of adultswas due towell-established evidence sug-gesting that cognitive abilities decline (e.g., Hasher, 1997; Rabbitt, 1993;

Table 2Performance scores (A, upper panel) and correlation with g (B, lower panel) across RIPresent and RI Absent groups in Experiment 1.

A RI Present(n = 38)

RI Absent(n = 31)

Comparison acrossgroups

Performance ANCOVA

M SD M SD F(1, 66) p ɳp2Critical error 0.65 0.25 0.34 0.34 18.51 b0.001 0.219Criterion fail 0.66 0.20 0.53 0.29 4.82 0.032 0.068Miss 0.26 0.15 0.20 0.15 2.16 0.146 0.032Hand error 0.08 0.08 0.12 0.11 4.46 0.038 0.063False positivek 0.08 0.18 0.02 0.03 2.81 0.098 0.041False positivev – – 2.71 1.75 – – –Reaction time (ms) 1201 187 1168 222 0.21 0.648 0.003

B Correlation with g Fisher's z-testr(36) p r(29) p z(67) p

Critical error 0.56 b0.001 0.00 0.999 2.50 0.012Criterion fail 0.76 b0.001 0.27 0.142 2.84 0.005Miss 0.52 b0.001 0.38 0.035 0.70 0.484Hand error 0.40 0.013 0.27 0.142 0.58 0.562False positivek 0.48 0.002 0.05 0.789 1.87 0.062False positivev – 0.27 0.142 – –Reaction time (ms) 0.24 0.147 0.48 0.006 1.10 0.271


Salthouse, 1992) and become more variable (e.g., Morse, 1993; Rabbitt,1993) in later years. Accordingly, older adult performancewas assumedto be more sensitive to the demands imposed by the task.

2.1.2. Tasks and procedure

2.1.2.1. Test of ‘g’: culture fair. Participants first completed Cattell's Test of‘g’: Culture Fair, Scale 2, Form A (Cattell, 1971; Cattell & Cattell, 1973;hereafter termed the Culture Fair). The Culture Fair loads highly onto gat r = 0.81 and comprises four subtests that measure novel problem-solving ability using geometrical figures in a set amount of time: seriescompletions, classifications, matrices, and topological relations.

2.1.2.2. Colour-shape match task. After completing the Culture Fair, par-ticipants performed the colour-shape match task, programmed usingPsyScope software (Cohen, MacWhinney, Flatt, & Provost, 1993). Thistask has previously been employed, in slightly modified form, byDuncan et al. (2008). Pairs of coloured outline shapes, one containinga tick in the centre and the other containing a cross, were presented inthe centre of a high resolution colourmonitor with awhite screen back-ground. Objects varied along two dimensions: colour (red, blue, green)and shape (circle, square, triangle). Across pairs, all combinations of col-our and shape were possible. Each object was 12.7 mm × 12.7 mm andthe distance between objects in a pair was 7.4 mm edge to edge; partic-ipants sat comfortably and viewed these stimuli without precise controlof viewing distance (approximately 0.5m). Pairswere amixture of non-matching (objects of different colour and shape), single-matching (ob-jects of same colour or shape), and double-matching (objects of samecolour and shape). In each trial, READYwas presented on screen, follow-ed by a blank-screen interval of 1500 ms, followed by 10 pairs whichwere presented for 1200 ms each with a 200 ms blank-screen intervalbetween each pair (see Fig. 1). Of the 10 pairs, three were single-matching, which were always two of the first five frames (with atleast one non-matching pair in between) and the ninth or 10th frame,one was double-matching (in 75% of trials), which was always the sev-enth frame, and the others were non-matching. Altogether, there were12 trials of 10 pairs. For the purpose of design, the 12 trials were splitinto three sub-blocks of four trials. In each sub-block, one trial hadonly colour-single-matching pairs, one trial had only shape-single-matching pairs, and two trials had a mixture of colour- and shape-single-matching pairs. Additionally, one trial required only left re-sponses, one trial required only right responses, and two trials requireda mixture of left and right responses. A double-matching pair was pres-ent in three of the four trials in each sub-block. Replication of the samestimulus pairs was restricted and did not occur within trials or sub-blocks.

The task required identification of pairs of items that shared thesame colour or the same shape. If an itempairmatched on either dimen-sion, the participant was to respond by pressing one of two responsekeys (with their right hand), contingent upon the position of the tick(left or right). Participants were then informed that, towards the endof each trial, they might see an item pair that matched in both colourand shape. The RI Present group was instructed to ignore thesedouble-matching frames whereas the RI Absent group was instructedto say “double” out loud in addition to pressing the appropriate re-sponse key. All key presses were recorded in PsyScope and were attrib-uted to a frame if they occurred within 2800 ms (two complete framesplus intervals) of its onset. The participants were not informed of thistime limit but simply that they were to respond as quickly but as accu-rately as possible. Participants initiallywent through an example trial onpaper with the experimenter and then completed a practice trial. Theywere then asked to repeat the rules and to answer a series of questions(e.g., “what do you do if both items of a pair share both colour andshape?”). If the ruleswere recalled or the questions answered incorrect-ly, the rules were repeated and the practice trial re-run. After comple-tion of the task, participants were asked to repeat the task rules.

2.1.3. DesignPerformance measures for the colour-shapematch task are present-

ed in Table 1. Key measures on the task were criterion fails, which tookinto account performance across task elements and were considered ameasure of overall success on a trial, critical error (RI Present), whichwas the ‘no-go’ measure due to its reliance on response inhibition, andmisses, which was the ‘go’ measure. The other measures presented inTable 1 have been included to illustrate performance across the fullrange of task components, but our primary focus here is on critical er-rors and criterion fails. All measureswere recorded as error as a propor-tion of all possible instances, except for false positivesv which wererecorded as raw values. All analyses were based on the full dataset(i.e., all 12 trials rather than separately by sub-block). Culture Fair rawerror scoreswere correlatedwith each performancemeasure separatelyfor the RI Absent and RI Present groups. Performance scores and corre-lations between those scores and Culture Fair error were comparedacross groups. The specific sequencing of trials varied pseudo randomlyand was maintained for all participants in order to hold the level of re-sponse prepotency constant.

2.2. Results

Table 2 presents performance scores (upper panel) and correlationswith g (as estimated by Culture Fair; lower panel). Analysis of covari-ance (ANCOVA; controlling for Culture Fair error) revealed that perfor-mance was significantly worse in the RI Present group relative to the RIAbsent group for critical errors (response to double-matching ‘no-go’pairs [RI Present] vs. missed response to double-matching pair [RI Ab-sent]; p b 0.001), criterion fails (overall success on a trial; p = 0.032)and hand errors (response to the incorrect side of single-matchingpair; p = 0.038). There was no significant difference across RI Presentand RI Absent groups for misses (missed response to single-matching‘go’ pair), false positives (response to non-matching ‘neutral’ pair), orreaction time (p ≥ 0.1 for all comparisons). Although therewas a generaltrend for reduced error and faster response time as participantsprogressed through the task, once a requirement had been satisfied par-ticipants commonly produced error on that requirement on later trials –and this was particularly the case for the critical double-matchingframes requiring no response. Twenty-nine of the 38 participants inthe RI Present group (76%) produced a critical error in trial 1, with thisfigure reducing to 19 (50%) on the final double-matching frame (trial11).


The majority of performance measures (all but reaction time) weresignificantly correlated with g (as estimated by Culture Fair raw errorscores) in the RI Present group, but only the miss–g and reactiontime–g correlations reached significance in the RI Absent group. Fisher'sz-test of difference for independent correlations revealed that the corre-lation with g was significantly stronger in the RI Present group relativeto the RI Absent group for critical errors, z(67) = 2.50, p = 0.012,with r(RI Absent) outside the 95% confidence interval (CI) [0.27, 0.75]for r(RI Present), calculated using Fisher's z transform. The criterionfail–g correlation was also significantly stronger in the RI Presentgroup, z(67) = 2.84, p = 0.005, with r(RI Absent) outside the 95% CI[0.58, 0.87] for r(RI Present). Therewas no significant difference formis-ses, hand errors, false positives, or reaction time (p N 0.2 in all compar-isons except hand errors which was marginal at p = 0.062). Fig. 2demonstrates these patterns of correlation, with more than twice asmany critical errors and criterion fails in the RI Present version of thetask produced by participants N 1 SDbelow the samplemeanon theCul-ture Fair in comparison to those N1 SD above the mean (upper panel).The absence of a meaningful relationship between these performancemeasures and Culture Fair in the RI Absent task variant is demonstratedin the lower panel, although the two participants N 1 SD above the Cul-ture Fair mean performed disproportionately better than the other par-ticipants in this group. All participants correctly repeated theinstructions at the end of testing.

2.3. Discussion

The inhibition (RI Present) taskwas clearly highly demanding. Itwasexpected that difficulties in withholding a response to the doublematches would be observed, particularly in low g participants, but thathigh g participants would be able to satisfy this requirement. In fact, vir-tually all participants were unable to prevent a response to these stim-uli. Nevertheless, remembering the task instructions was not difficult;there were a few, well-specified rules and no obvious element ofproblem-solving. All the participants correctly repeated the task rulesonce the experiment had ended, demonstrating that they did not forgetwhat was required of them. Moreover, participants typically reactedverbally immediately upon producing a response to the doublematching frames, indicating that the demand to inhibit the responsewas in mind but the strength of two response tendencies (i.e., colourmatch and shape match) was such that the instruction was overturnedand the key inadvertently pressed. Under the considerable task de-mands the observed behaviour emulates that sometimes described inpatients with frontal lobe damage: production of erroneous behaviourdespite simultaneous correct verbalisation of task rules (Luria, 1966,1973).

Performance on critical double-matching items was significantlycorrelated with g when a strong response inhibition demand was re-quired, but the relationshipwas negligiblewhen replaced by an alterna-tive positive response. Additionally, performance across all taskconstraints was more closely correlated with g when the taskencompassed the response inhibition demand, indicating that difficul-ties in resolving response conflicts are particularly common in individ-uals at the lower end of the g distribution. The findings raise theprospect that a requirement to inhibit prepotent response tendenciesmay be a fundamental “risk factor” for the recruitment of g in task per-formance. Nevertheless, because, on average, performance was betteron the RI Absent task, the difference in g-correlations across tasks maybe driven by broader issues associated with differences in overall taskdifficulty.We address this issue in Experiment 3, following a replicationof our RI Present task in a younger group of participants.

3. Experiment 2: a replication in younger participants

Experiment 1 was conducted with older participants that, on aver-age, scored 0.5 SDs below published adult Culture Fair norms, and

there was a comparative absence of those scoring at the high end ofthe normative distribution. Variability in performance is known to in-crease with age (Rabbitt, 1993), desirable to the extent that a broadrange of scores may be collected. A negative consequence, however, isthat different cognitive abilities may show differential sensitivities toage (e.g., Kievit et al., 2014) and it cannot be assumed that similar pat-terns of performancewill be observed across different age groups. In Ex-periment 2, we therefore assessed the sensitivity of performance on theresponse inhibition task to Spearman's g in a younger group of partici-pants. It was predicted that the level of error would be lower thanthat observed in Experiment 1, but performance would remain highlycorrelated with Culture Fair scores.

3.1. Method

3.1.1. ParticipantsThirty neurologically healthy adults (7 males, 23 females), aged be-

tween 30 and 50 years (M = 40.7, SD = 6.1), were recruited from theMRC Cognition and Brain Sciences Unit panel of volunteers. CultureFair error (M=12.43 SD=5.27) was significantly lower than in Exper-iment 1 (RI Present), a difference of approximately 10 IQ points, usingnormalised standard scores (Cattell & Cattell, 1973). All participantsgave informed consent and were paid to participate.

3.1.2. Tasks and procedureThe task instructions, procedure and performance measures were

identical to those administered to the RI Present group in Experiment 1.

3.2. Results

Performancewas significantly better in comparison to Experiment 1on most measures—including the key measures, that is, misses (missedresponse to single-matching ‘go’ pair), critical errors (inappropriate re-sponse to double-matching ‘no-go’ pair) and criterion fails (overall suc-cess on a trial)—although reaction times were closely similar (Table 3,upper panel). However, when controlling for Culture Fair error, therewas a convergence of performance across experiments such that nocomparisons reached statistical significance (at p=0.05). Thus, the per-formance differences between the younger (Experiment 2) and older(Experiment 1) groups appear largely attributable to fluid intelligence(asmeasured by Culture Fair). The proportion of participants producinga critical error on trial 1 (73%) was similar to that observed in Experi-ment 1 (76%), although there was a steeper practice effect across trials(only 13% of participants produced a critical error on the final doublematching frame, compared to 50% of participants in Experiment 1).

Correlations of performance against Culture Fair were numericallysmaller than those observed in Experiment 1 (Table 3, lower panel).Nevertheless, direct comparison of coefficients across experiments(using Fisher's z-test of difference for independent correlations) revealedno statistically significant differences in the strength of these correla-tions. The correlations were significant for critical error and criterionfails. Fig. 3 presents a comparison of performance in younger (Experi-ment 2) against older (Experiment 1) participants, grouped by CultureFair z score bins of width 0.5 SD. Despite better Culture Fair performance(with a complete absence of participants in the lowest bin), the patternof performance in younger participants was similar to the older group,with those scoring above the Culture Fairmeanproducing disproportion-ately fewer critical errors and criterion fails than those below the mean.

3.3. Discussion

Results from Experiment 2 indicate that the sensitivity of the inhibi-tion (RI Present) task to variations in Culture Fair performance is robustand operates in younger, as well as older, participants. Close correspon-dence was observed across experiments in the proportion of critical er-rors and criterion fails produced when performance was pooled by

Fig. 2. Critical error and criterion fail in the RI Present and RI Absent groups across Culture Fair z score bins in Experiment 1. Error bars represent standard error for data within therespective bin. The numbers in the individual columns represent number of participants in that z score bin.


Culture Fair score. Indeed, once Culture Fair performance was partialledout, there were no reliable differences in level of performance betweenthe younger and older groups.

Thefindings fromExperiments 1 and 2 indicate that difficulties in re-solving response conflict operate in the normal population, and are par-ticularly common in individuals at the lower end of the g distribution.However, using the criterion of errors produced, the RI Present taskwas clearly more demanding than the RI Absent task, and this raisesthe possibility that the strong performance–g correlations were drivenby more general demands associated with increased task difficulty orcomplexity rather than by the demand for response inhibition. Thisissue is addressed in Experiment 3.

4. Experiment 3: response inhibition versus task complexity

In Experiment 3we focused on real-time processing demands, aswedid in Experiments 1 and 2, examining the role of task complexity in thecontext of response inhibition. From a limited attentional capacity per-spective, selectively attending to one or more task demand will reducethe capacity for attending to additional demands (e.g., Broadbent,1958; Desimone & Duncan, 1995). Consequently, the more task con-straints there are, the more likely it is that one of these will fail tobe selected or acted upon. In Experiment 3, therefore, we examinewhether the association of task performance with g, demonstratedin Experiments 1 and 2, can be explained primarily on the basis of

Table 3Performance scores (A, upper panel) and correlation with g (B, lower panel) across younger (Experiment 2) and older (Experiment 1) groups.

A Younger (n = 30) Older (n = 38) Comparison across groups

Performance ANOVA ANCOVA

M SD M SD F(1, 66) p ɳp2 F(1, 65) p ɳp2Critical error 0.44 0.30 0.65 0.25 9.80 0.003 0.129 1.41 0.239 0.021Criterion fail 0.46 0.20 0.66 0.20 13.99 b0.001 0.175 1.88 0.175 0.028Miss 0.13 0.14 0.26 0.15 12.86 0.001 0.163 3.24 0.077 0.047Hand error 0.03 0.04 0.08 0.08 8.42 0.005 0.113 1.91 0.172 0.029False positivek 0.02 0.01 0.08 0.18 3.03 0.086 0.044 0.00 0.950 0.000Reaction time (ms) 1180 140 1201 187 0.27 0.605 0.004 0.22 0.641 0.003

B Correlation with g Fisher's z-testr(28) p r(36) p z(66) p

Critical error 0.46 0.014 0.56 b0.001 0.53 0.596Criterion fail 0.54 0.003 0.76 b0.001 1.72 0.126Miss 0.32 0.097 0.52 b0.001 0.96 0.337Hand error 0.23 0.239 0.40 0.013 0.74 0.459False positivek 0.18 0.359 0.48 0.002 1.33 0.184Reaction time (ms) 0.32 0.097 0.24 0.147 0.34 0.734


task complexity (measured by the number of task constraints) orwhether there is something fundamental about response inhibitionin the performance-g relationship. In order to address this issue wemanipulated the tasks such that the complexity of the RI Absenttask was increased and the complexity of the RI Present task was re-duced. In the Simple RI Present task, the requirement to press left orright was removed, leaving only one response key for single-matching frames. In the Complex RI Absent task, items varied alongthree dimensions: colour, shape, and shading. We predicted that al-though the Simple RI Present task would be easier to perform thanthe RI Present task of Experiments 1 and 2, sensitivity to variationsin fluid intelligence would remain. We also predicted that moderatecorrelations between performance and gwould be found in the Com-plex RI Absent task, due to increased attentional demands, but thatthese associations would not exceed those observed in Simple RIPresent, despite the likelihood of higher overall level of error. Thisresult would suggest that the role of g in task performance increaseswith task complexity but that a process of action control via suppres-sion of inappropriate response tendencies may be a fundamental riskfactor for the recruitment of g that operates relatively independentlyof task complexity.

Fig. 3. Critical error and criterion fail in the younger (Experiment 2) and older (Experiment 1) grstandard error. The numbers in the individual columns represent the number of participants in

4.1. Method

4.1.1. ParticipantsParticipants were 36 neurologically healthy adults (7 males, 29 fe-

males) aged between 22 and 52 years (M=38.5, SD=9.5); mean Cul-ture Fair error was 11.92 (SD = 4.87). All participants were recruitedfrom the MRC Cognition and Brain Sciences Unit paid panel of volun-teers and each participant gave informed consent.

4.1.2. Tasks and procedure

4.1.2.1. Simple RI Present colour-shape match task. The task and instruc-tionswere identical to those administered to the RI Present group of Ex-periment 1 with the exception that the ticks and crosses were removedfrom the centre of the objects and that participants were required topress one button for all single-matching pairs (see Fig. 4, left panel).

4.1.2.2. Complex RI Absent colour-shape match task. The task and instruc-tions were identical to those administered to the RI Absent group of Ex-periment 1with the following exceptions: a) the ticks and crosses in thecentre of the objects were replaced by a horizontal line which indicated

oups across Culture Fair z score bins. Charts showRI Present data only. Error bars representthat z score bin.

Fig. 4. Representation of typical Simple RI Present and Complex RI Absent trials in thecolour-shape match task. Black, mid grey, and light grey objects are presented in blue,red, and green, respectively. (For interpretation of the references to colour in this figurelegend, the reader is referred to the web version of this article.)

Table 4Performance scores (A, upper panel) and correlation with g (B, lower panel) for Simple RIPresent and Complex RI Absent tasks in Experiment 3.

A Simple RI Present Complex RI Absent

Performance

M SD M SD

Critical error 0.08 0.07 0.53 0.30Critical error (trial 1) 0.28 0.45 – –Criterion Fail 0.11 0.12 – –Misscs 0.02 0.04 0.17 0.15Misssh – – 0.70 0.25Within hand error – – 0.14 0.13Between hand error – – 0.10 0.13False positivesk 0.004 0.01 0.03 0.11False positivesv – – 2.14 4.86Perfect trials – – 0.09 0.18Neglected requirements – – 0.28 0.57Reaction time (ms) 844 96 1258 175

B Correlation with gr(34) p r(34) p

Critical error 0.44 0.007 0.40 0.016Critical error (trial 1) 0.42 0.011 –Criterion Fail 0.24 0.159 –Misscs 0.13 0.450 0.13 0.450Misssh – 0.47 0.004Within hand error – 0.01 0.954Between hand error – 0.08 0.643False positivesk 0.02 0.908 −0.23 0.177False positivesv – −0.08 0.643Neglected requirements – 0.38 0.022Reaction time (ms) 0.34 0.042 0.10 0.562

Note. Misssh = no response to colour or shape single-matching (GO) pairs; Misssh = noresponse to shade single-matching pairs; False positivesk = response (key press) to non-matching pairs; False positivesv= response (verbal) to single or non-matching pairs; n=36.


the required response (left or right)with the right hand as shown in Fig.4 (right panel); b) pairs varied across three levels of shading (as well asacross colour and shape). In 75% of trials one pair matched in shade,which also required a response (with the left hand) according to the po-sition of the horizontal line. A shadematch occurred only in frames thatdid not match in colour or shape. Thus, participants were required topress the appropriate key (left or right) when items matched on colourand/or shape andwhen itemsmatched on shade. For the 75% of trials inwhich the seventh frame matched on both colour and shape, partici-pants were required to press the relevant key and say “double” (aswas the case for the RI Absent group in Experiment 1).

Participants completed the two versions of the colour-shape matchtask—Simple RI Present and Complex RI Absent—with the Culture Fair(Scale 2, Form A) being administered in between the two task variants.Instructions were given in the same manner as in Experiment 1, withthe removal/addition of performance requirements to reflect the de-mands of these task variants. For each task, participants initially workedthrough an example trial on paper with the experimenter and complet-ed a practice trial, and were then asked to repeat the rules. If recalled

incorrectly, the rules were repeated and the practice trial was re-run.The experimental trials, therefore, did not proceed until clear under-standing of the instructions had been demonstrated. Participants werealso asked to recall the instructions after completion of each task.

4.1.3. DesignPerformance measures were the same as those in Experiment 1,

with the addition, in the Complex RI Absent task, of misssh (missed re-sponse to single-matching items matching in shade), between handerror (response with the wrong hand), neglected requirements (com-plete failure to attend to one or more requirements throughout thecourse of the task), and perfect trials (satisfaction of all task demandson individual trials). False positivesv and neglected requirements wererecorded as raw values whereas the remainder of performance mea-sureswere recorded as proportion of error. Performance scores and cor-relations between performance scores and Culture Fair error werecompared across the Simple RI Present andComplexRI Absent tasks. Or-dering of the two experimental tasks was counterbalanced, with half ofthe participants completing the Simple RI Present task first.

4.2. Results

Performance scores (upper panel) and correlations with g (lowerpanel) are presented in Table 4. On the Simple RI Present task, erroron all measures was low, with participants on average making fewerthan one miss (missed response to single-matching ‘go’ pair), criticalerror (inappropriate response to double-matching ‘no-go’ pair), orfalse positive (response to non-matching ‘neutral’ pair) across all 12 tri-als. However, over 25% of participants failed to suppress a response tothe double-matching frame in trial 1. Overall, a mean of 11% of trialswere failed. On the Complex RI Absent task, though, a substantial pro-portion of error was found on all measures. Participants failed to re-spond verbally to over 50% of double-matching items (critical errors)


and missed over two-thirds of items matching on shade (misssh). Justover one of the 12 trials was completed without error (perfect trials)and a minority of participants showed a complete failure to attend toone or more requirements throughout the course of the experiment(neglected requirements). Nearly 20% of single-matching colour/shapeitems were missed. Of the targets that were responded to, an incorrectkey was commonly pressed (14% with the correct hand and 10% withthe incorrect hand). False positives occurred infrequently, both interms of erroneous key responses and verbal responses to non-matching items.

Correlations with g were typically lower when the RI Present taskwas simplified, but remained highly significant for critical errors despitebetter performance on this measure (and all other measures). Signifi-cant correlations emerged in the Complex RI Absent task. Fig. 5 demon-strates the trend for better performance in participants at the higherend of the g distribution. On the Simple RI Present task, all 10 partici-pants scoring N1 SD above the mean on the Culture Fair were able towithhold responding to double-matching items on thefirst trial, where-as of those scoring N1 SD below the population mean on the CultureFair, over 50% produced a critical error. The chart of criterion fails alsoclearly indicates a pattern of improved performance with increasingCulture Fair score; that this correlation did not reach significance ap-pears primarily due to performance in those participants in the lowestz score bin (a correlation of r = 0.37, p = 0.044 was observed whenomitting those participants from the analysis). On the Complex RI

Fig. 5. Critical error and criterion fail in the simple RI Present task (upper panel) andmisses (shaFair z score bins in Experiment 3. Error bars represent standard error. The numbers in the indi

Absent task, neglected requirements were largely restricted to partici-pants at the low end of the g distribution. Other significant correlationsalso emerged, including for perfect trials, as indicated in Table 4. Instruc-tions were recalled correctly at the end of testing.

4.3. Discussion

Experiment 3 demonstrates that complexity, or the number of con-current task demands, plays an important part in the involvement ofSpearman's g in task performance. The Simple RI Present task was asso-ciated withmoremoderate correlationswith g thanwas foundwith thestandard RI Present task employed for Experiment 1, although the cor-relations were more consistent with Experiment 2. This result impliesthat, while the reversal demand may represent an important part ofthat association, as indicated by the sensitivity of critical errors to Cul-ture Fair performance, other factors may have a mediating influence.The reintroduction of a relationship between error and Culture Fair inthe Complex RI Absent task suggests that the number of requirementsthat must be concurrently attended to partly determines the strengthof the association. The observation that elements of performance onan easier version of the RI Present task predicted Culture Fair scores sug-gests that a process of action control via management of response po-tency may be a basic process underlying g. However, involvement of gin this type of response suppression is complicated by more general is-sues of task complexity.

de), critical error and neglected requirements in the complex RI Absent task across Culturevidual columns represent the number of participants in that z score bin.

Table 5Task instruction rules for Experiment 4.

Group Rule Description

Two Rule Rule 1 Respond to items that match in colour OR in shape bypressing the side corresponding to the tick

Rule 2 Ignore items that do not match in colour or shape, anditems that match in both colour AND shape

Four Rule Rule 1 Respond to items that match in colour by pressing theside corresponding to the tick

Rule 2 Respond to items that match in shape by pressing theside corresponding to the tick

Rule 3 Ignore items that do not match in colour or shapeRule 4 Ignore items that match in both colour AND shape


In the Complex RI Absent task, error was most readily observed onthe shade requirement, which was specified last. That the order of in-structions is important in predicting which demand will be failed can-not be stated with any certainty, as all participants received the sameorder; the shade demand may simply be more difficult to satisfy thanthe others. However, the finding is consistent with the suggestion of aprimacy effect, whereby requirements specified early are themost like-ly to be satisfied (Duncan et al., 1996). Of the various task requirements,this was also the onemost sensitive to variations in g (although not sig-nificantly so). This finding is consistent with a reduced WMC explana-tion for individual differences in performance.

Itemsmatching on colour and shape (in both tasks) and on shade (inComplex RI Absent) drove the g-correlations observed in this experi-ment, suggesting that frequency of occurrence may also be importantin determining relative success on demanding tasks. The requirementto respond to frames matching on colour or shape was reinforcedmore frequently than other demands, therefore affecting a responsebias towards these frames. In every trial, there were three framesmatching on colour or shape, with a maximum of one shade matchand one colour and shape match per trial. It is intuitive that a task de-mand is less likely to lose control of behaviour if it is being regularly re-inforced by stimuli matching that demand. Our findings indicate thatvariability in the capacity for effectively counteracting a response biasin line with task goals is associated with individual variability in g.

5. Experiment 4: response inhibition versus task instructioncomplexity

In Experiment 4 we focus on the complexity of task instructions toaddress recent claims that the total complexity of task instructions(the “task model”), rather than real-time task execution demands, aremost closely associated with measures of Spearman's g (Dumontheilet al., 2010; Duncan et al., 2008; Duncan et al., 2012; Roberts &Anderson, 2014). The claims were based on a series of tasks in whichcomplexity was manipulated by varying the task instructions, whichcould be presented as a full or a reduced set (Duncan et al. 2008). Inthe full instructions condition requirements for two tasks were ex-plained, and the participants were told that information for one of thetasks could be temporarily discarded. In the reduced instructions condi-tions, requirements for one task were explained, and after completion,requirements for the second task were explained. We have applied adifferent approach in which we manipulated WM load by splitting thesame executable task information into two or four distinct chunks ofinformation.

Participants completed the RI Present task of Experiments 1 and 2,and an RI Absent task which was operationally identical to the RI Pres-ent task except that double-matchingpairswere never actually present-ed. One group of participants—the Two Rule group—were given taskinstructions chunked into two rules (assumed to represent an efficientmental representation of the task), and another group—the Four Rulegroup-–were given the same instructions instead chunked into fourrules (assumed to represent an inefficient task representation). This ap-proach provided the opportunity for investigating a) the importance ofresponse inhibition in the context of efficiently and inefficiently con-structed task instructions (while holding the actual operative task relat-ed information constant), and b) the effects of modelling taskrequirements which are not required during actual task execution. Wepredicted that performance–g correlations would be stronger for four-rule relative to two-rule instructions, despite holding constant informa-tion directly related to the task as well as the task execution demandsthemselves. If efficiency of taskmodelling, rather than execution, is fun-damental to Spearman's gwemight also expect tofind evidence that re-assembly of task rules, towards greater efficiency, would be restricted tothose participants at the higher end of the g distribution. Ifperformance–g correlations remain strong across both RI Present andRI Absent tasks in the four-rule condition, this would bemost consistent

with the claim that task modelling rather than task execution is themore fundamental risk factor for the recruitment of g.

5.1. Method

5.1.1. ParticipantsOne-hundred neurologically healthy adults (37 males, 63 females),

aged between 18 and 63 years, were recruited from the Departmentof Psychology, Anglia Ruskin University in Cambridge and the widerlocal community. Psychology undergraduates were recruited via anon-line study recruitment systemwhich granted course credit; commu-nity volunteers were recruited via word-of-mouth and did not receiveany payment. None of the participants were carried over from previousexperiments. The Two Rule group comprised 50 adults (20males, 30 fe-males) aged between 18 and 62 years (M = 30.98, SD = 12.56). TheFour Rule group comprised 50 adults (17 males, 33 females) aged be-tween 18 and 63 years (M = 29.22, SD = 12.78). Culture Fair errorwas equivalent across the Two Rule (M = 9.98, SD = 4.09) and FourRule (M = 11.54, SD = 5.12) groups, t(98) = 1.68, p = 0.096, d =0.339. As in previous experiments, the sample sizes are similar tothose of studies addressing similar research questions (e.g., Duncanet al. 2008; Roberts & Anderson, 2014).

5.1.2. Tasks and procedureParticipants completed the Culture Fair (Scale 2, Form A) followed

by the two versions of the colour-shape match task in succession. TheRI Present task variant was identical to the 12 trials of Experiment 1(see Fig. 1); the RI Absent task variant was identical to RI Present excepttheninedouble-matching frameswere replaced bynine additional non-matching frames. The instructions of previous experiments were modi-fied here. Participants were informed that pairs of coloured objects, onecontaining a tick and the other containing a cross, would be presentedone at a time in quick succession. They were informed that objectscould share either the same colour or the same shape and that towardsthe end of each sequence, they might see a pair that shares both thesame colour and the same shape. Half of participants were then giventwo rules to follow and the other half of participants were given fourrules to follow; essentially the information contained across rule for-mats was the same, but was presented in the form of either two orfour chunks (see Table 5). Participants worked through an exampletrial with the experimenter as in previous experiments but here therewas no practice trial. The rules were then repeated for a second timeand the participant was asked to repeat the rules (which were record-ed). Unlike previous experiments, participants were not asked a seriesof questions about the task. If parts of the ruleswere omitted, the exper-imenter repeated the appropriate rule until the participant was able torepeat all parts of all rules. This ensured that task instructions were ef-fectively stored in short-term memory such that every participantknew what to do. Immediately before each task, participants were toldthat the presentation of items would be very fast, to not be surprisedat this, to just do their best, and that they would get a chance for ashort break after every 10th pair of items. Immediately after each

Table 7Paired t-tests comparing performance scores between RI Present and RI Absent tasks (A,upper panel) and between critical errors and other measures (B, lower panel) in Experi-ment 4.

A Two Rule Four Rule

RI Present vs. RI Absent

t(49) p Cohen's d t(49) p Cohen's d

Criterion fail 9.75 b0.001 1.369 10.48 b 0.001 1.669w/o critical error 4.17 b0.001 0.560 4.33 b 0.001 0.682

Miss 3.87 b0.001 0.503 4.68 b0.001 0.743Hand error 0.92 0.362 0.238 0.32 0.750 0.177False positivek 1.59 0.118 0.223 0.49 0.626 0.127Reaction time (ms) 3.90 b0.001 0.563 5.32 b 0.001 0.757

B Critical error vs. other measurest(49) p Cohen's d t(49) p Cohen's d

RI PresentMiss 4.20 b0.001 0.648 5.86 b0.001 1.095Hand error 9.55 b0.001 1.760 10.67 b0.001 2.233False positivek 6.84 b0.001 1.026 10.71 b0.001 1.249

RI AbsentMiss 9.66 b0.001 1.491 10.40 b0.001 1.835Hand error 5.93 b0.001 0.708 7.94 b0.001 1.245False positivek 8.23 b0.001 1.895 11.47 b0.001 2.231


version of the task, participants were asked to repeat the rules as theyremembered them. Key presses were attributed to a frame if they oc-curred within 1200 ms of its onset.

5.1.3. DesignPerformance measures were again identical to Experiment 1 (RI

Present). All performance measures were recorded as proportion oferror. Performance scores and correlations between performance scoresand Culture Fair error were compared across RI Absent and RI Presenttask variants and across the Two Rule and Four Rule groups. The orderof task variants was counterbalanced across participants, with half ofparticipants in each group completing the RI Present task first.

5.2. Results

Performance scores were strikingly similar across the Two Rule andFour Rule groups, as displayed in Table 6; ANCOVA showed that the onlyperformance measure that differed across rule groups was false posi-tives in RI Present (response to non-matching ‘neutral’ pair; p = 0.04).Paired t-tests (see Table 7) confirmed that, in each rule group, errorwas significantly greater for critical errors than for any other measure(excluding criterion fails, i.e., overall success on a trial; all p b 0.001, sig-nificant using the Bonferroni-corrected alpha level of p b 0.05/6 =0.008). There was a trend for greater error and slower reaction timesin RI Present compared to RI Absent for every performance measure;in each rule group, this effect was significant for misses (missed re-sponse to single-matching ‘go’ pair), reaction time, and criterion fails(when both including and excluding critical error as a criterion fortrial failure; all p b 0.001, significant using the Bonferroni-correctedalpha level of p b 0.05/6= 0.008; see Table 7). The proportion of partic-ipants failing to withhold a response to critical double-matching frames(critical error) decreased from 32 to 15% (in Two Rule) and from 54 to32% (in Four Rule) across the 12 trials. Again, successful inhibition didnot prevent later error on critical items; this was true for 78% of theTwo Rule group and for 80% of the Four Rule group.

Performance–g correlations across rule groups (Two Rule vs. FourRule) are presented in Table 8. Fisher's z-tests showed that the criticalerror–g correlation was stronger in the Four Rule group, with r(TwoRule) outside the 95% CI [0.31, 0.71] for r(Four Rule), although this dif-ference was marginally outside the statistical threshold of p = 0.05(p=0.051, two-tailed). The criterion fail–g correlationwas significantlystronger in Four Rule, with r(Two Rule) outside the 95% CI [0.45, 0.79]for r(Four Rule). There was a trend for higher performance–g correla-tions a) for critical errors relative to othermeasures, and b) in RI Present

Table 6Performance scores across Two Rule and Four Rule groups in the RI Present (A, upper pan-el) and RI Absent (B, lower panel) tasks in Experiment 4.

A Two Rule (n =50)

Four Rule (n =50)

Comparison acrossgroups

RI Present ANCOVA

M SD M SD F(1, 97) p ɳp2Critical error 0.40 0.26 0.43 0.27 0.00 0.947 0.000Criterion fail 0.50 0.27 0.49 0.27 1.01 0.317 0.010w/o critical error 0.30 0.28 0.28 0.23 1.43 0.235 0.015Miss 0.20 0.16 0.20 0.16 0.37 0.546 0.004Hand error 0.04 0.04 0.06 0.09 0.64 0.426 0.007False positivek 0.08 0.19 0.03 0.05 4.44 0.037 0.044Reaction time (ms) 850 87 841 93 0.22 0.639 0.002

B RI Absent ANCOVAM SD M SD F(1, 97) p ɳp2

Criterion fail 0.18 0.24 0.14 0.13 1.93 0.168 0.020Miss 0.13 0.14 0.11 0.09 1.57 0.213 0.016Hand error 0.03 0.04 0.05 0.08 1.67 0.199 0.017False positivek 0.05 0.17 0.02 0.06 2.10 0.151 0.021Reaction time (ms) 816 95 792 90 1.68 0.198 0.017

relative to RI Absent, but only in the Four Rule group. Williams-Hotelling t-tests revealed that the critical error–g correlation did not dif-fer significantly from other performance–g correlations in Two Rule (allp N 0.1), but in Four Rule, the correlation with gwas significantly higherfor critical errors relative to false positives in RI Present, t(47) = 2.22,p=0.03, reaction time in RI Present, t(47)= 3.26, p=0.002, and reac-tion time in RI Absent, t(47) = 3.15, p = 0.001. These latter two find-ings, however, would be expected due to negligible correlations forresponse time in each block. Additionally, performance–g correlationsdid not differ significantly across RI Present and RI Absent task variantsfor any measure in the Two Rule group (all p N 0.1), but the RI Presentcriterion fail–g correlation was significantly stronger in RI Present rela-tive to RI Absent in the Four Rule group, t(47) = 2.32, p = 0.02.

Fig. 6 displays performance scores for critical errors and criterionfails across Culture Fair z-score bins. All charts demonstrate the strongerrelationship between task performance and g in four-rule relative totwo-rule conditions, and in RI Present relative to RI Absent task variants.For example, there was a 20% difference in critical error between lowerand higher g participants (those scoring N 1 SD below and above themean, respectively) in Two Rule, compared to a 34% difference in Four

Table 8Performance–g correlations across TwoRule and Four Rule groups in the RI Present (A, up-per panel) and RI Absent (B, lower panel) tasks in Experiment 4.

A Two Rule Four Rule Comparisonacross groups

RI Present Fisher's z-test

r(48) p r(48) p z(98) p

Critical error 0.20 0.164 0.54 b0.001 1.95 0.051Criterion fail 0.30 0.034 0.65 b0.001 2.26 0.024

w/o critical error 0.22 0.125 0.54 b0.001 1.84 0.066Miss 0.25 0.080 0.36 0.010 0.59 0.555Hand error 0.33* 0.019 0.40 0.004 0.39 0.697False positivek 0.14 0.332 0.17 0.238 0.15 0.881Reaction time (ms) −0.06 0.679 0.00 0.999 −0.29 0.772

B RI Absent Fisher's z-testr(48) p r(48) p z(98) p

Criterion fail 0.24 ,093 0.37 0.008 0.70 0.484Miss 0.24 0.093 0.21 0.143 0.15 0.881Hand error 0.30 0.034 0.33 0.019 0.16 0.873False positivek 0.19 0.186 0.32 0.023 0.68 0.497Reaction time (ms) 0.05 0.730 −0.02 0.890 0.34 0.734

Fig. 6. Critical error and criterion fail in the Two Rule and Four Rule groups across Culture Fair z score bins in Experiment 4. Error bars represent standard error. The numbers in theindividual columns represent the number of participants in that z score bin.


Rule. Criterion fails (incorporating critical error) varied from 28 to 19%(in Two Rule) and from 46 to 8% (in Four Rule) across lower and higherg participants, respectively, in the RI Present block. We also observeddifferences in the g-correlations between participants that completedthe RI Present task variant first and participants that completed the RIAbsent task first in the Four Rule group only. Fisher's z-test (two-tailed)revealed that the correlation with Culture Fair error was significantlystronger in participants (in the Four Rule group) completing the RI Ab-sent task first relative to participants completing the RI Present task firstfor the following measures: critical errors, z(48) = 2.23, p b 0.05, handerrors in RI Present, z(48)= 3.01, p b 0.01, and hand errors in RI Absent,z(48) = 2.78, p b 0.01.

When asked to repeat the rules at the end of testing, participantscould describe what was required of them during task execution.Some participants (Two Rule, n = 8; Four Rule, n = 19) stated therules in a different number of chunks than specified in their adminis-tered task instructions. Although no significant differences were ob-served, Culture Fair error tended to be lower in those participants thatreconceptualised the task towards greater efficiency (i.e., from tworules to one: M = 7.00 errors; or from four rules to three, two, or one:M = 10.84) relative to participants that did not reconceptualise therules (Two Rule:M = 10.17 errors; Four Rule: M = 10.93).

5.3. Discussion

The results from Experiment 4 suggest that regardless of whethertask instructions were administered as two or four rules, the demandto inhibit an inappropriate prepotent response tendency was relativelydifficult. In both rule groups, accuracy for critical double-matchingitems was significantly poorer than accuracy for any other measure,with the exception of some criterion fail measures which took into ac-count performance across task elements. This is in linewith predictions,

and with previous research showing that a) no-go trial performance isless accurate than go trial performance (e.g., Redick et al., 2011), andb) items specified last in instructions are more likely to be neglectedthan earlier specified items (Duncan et al., 1996). However, perfor-mance differed from the “goal neglect” reported in Duncan et al.(1996) in which once a requirement was satisfied it was (usually) notneglected again. Here, for the majority of participants successful inhibi-tion of a response to no-go items did not prevent later critical error. Per-formance on the task that required response inhibition was alsorelatively difficult for participants irrespective of how the instructionswere administered. Separately in each rule group, accuracy (misses)and speed (reaction time) for go items and overall performance (crite-rion fails) was significantly poorer in RI Present relative to RI Absent.Whether this was due to the inclusion of response inhibition demandper se, or was the consequence of RI Present having an additional taskconstraint, is unclear. Performance was so similar across rule groupsthat only one of the performancemeasures (false positives) differed sig-nificantly between them.

Differences between participants given task instructions as two rulesand participants given task instructions as four rules did, however,emerge in the correlational analyses. In linewith predictions, the correla-tionwithCulture Fair errorwas significantly stronger in Four Rule relativeto Two Rule for the two measures central to our hypotheses—critical er-rors and criterion fails in RI Present—and was numerically higher forthe majority of other measures. In as much as the four rule instructionsrepresented a more complex set of instructions, this is supportive ofother work addressing the relationship between fluid intelligence andtask complexity via manipulation of task instructions of a feature matchtask. In Duncan et al. (2008; adults) and Roberts and Anderson (2014;children) performance–g correlations were numerically stronger in thegroup administered an additional task rule even though this was actuallyredundant in task execution relative to the group given instructions in


which this unnecessary rule was omitted. Both of these studies also re-ported more error, reflected in a greater amount of rule neglect, in thehigh-complexity instruction group. Our observation of the emergenceof statistically stronger performance–g correlations in the high-complexity instruction group was, conversely, in the absence of poorerperformance in this group. This could be explained by methodologicaldifferences in manipulating task instruction complexity, with bothgroups in our study receiving the same task executable information, albe-it in different formats (two or four ‘chunks’). Given this observation ofequal performance in the two- and four-rule groups, the present findingscannot straightforwardly be explained with reference to task executiondifficulty per se.

We also found that response inhibition was important to recruit-ment of g, but only under four-rule instructions. The critical error–g cor-relation was significant in Four Rule (and not in Two Rule, with thedifference in strength between these correlations being just short of sig-nificance at p=0.051), and was also shown to be significantly strongerthan g-correlations for other performance measures (i.e., false positivesin RI Present and reaction time in both task variants) only in Four Rule.The non-significant critical error–g correlation found in our Two Rulegroup supports previous work that argues against a link between re-sponse inhibition and intelligence (e.g., Friedman et al., 2006), but thesignificant correlation in the Four Rule group supports previous workarguing for such a relationship (e.g., Salthouse et al., 2003). This leadsus to question whether conflicting results in the literature concerningthe relationship between inhibition and intelligence exist, at least inpart, because the complexity of task instructions, or the load imposedon theWMsystem by task instructions, has not been controlled. Indeed,evidence from task switching experiments suggests that switch costsoccur when task instructions elicit task sets but not when task instruc-tions suggest only stimulus-response associations (and also that whentask sets are known, they dominate performance relative to simplestimulus-response mappings, Dreisbach, Goschke, & Haider, 2007). An-other interesting observationwas that, when asked to state task rules atthe end of testing, some participants appeared to chunk task instruc-tions into fewer or more rules than were initially specified to them attask instructions. Further, a trend for fewer Culture Fair errors was ob-served in those participants that stated task rules as fewer chunks,and perhaps more efficiently. It is possible, however, that because par-ticipants received no explicit instruction to impose their own order onthe task rules, other instances of reconceptualisation were not identi-fied, perhaps masking any potential significant effects.

Differences between order groups were also observed only in theFour Rule group. As well as other measures (hand errors in both taskvariants) the correlation with Culture Fair for items requiring responseinhibition was significantly stronger for participants completing the RIAbsent task first. The requirement to ignore double-matching frameswas presented as a part of a rule (Rule 2) in Two Rule instructions butin isolation (Rule 4) in Four Rule instructions. If the ability to use or ex-ecute task rules is related to g then this findingmay be explained by thegreater ability of higher g participants to adhere to a rule that had notpreviously been used for correct performance. That is, by completingthe 12 trials of RI Absent (i.e., a task not containing double-matchingitems) prior to attempting RI Present, the response inhibition require-ment had not been enforced for Absent–Present participants. In con-trast, Present–Absent participants use, and therefore enforce, thedouble-match requirement straight away. This finding suggests greaterdemand for operating/executing a task rule that had not been enforcedduring previous performance of the task. Roberts and Anderson (2014)concluded that their additional unnecessary task instruction may haveincreased goal neglect by disrupting the conversion of task rules intosufficiently activated goal representations. Our results suggest thatreal-time task experience may act to increase activation level of goalrepresentations via enforcement of representations that are commonlyused. Collectively, the findings of Experiment 4 suggest that the way inwhich a task is cognitivelymodelled is critical to the involvement of g in

performance of that task, particularly under conditions of response con-flict. Perhaps whenWM capacity is stretched, additional recruitment ofg is required to manage the execution demands of the task.

6. General discussion

Our approach to investigating individual differences in task perfor-mance has been driven in large part byneuropsychological observationsof mismatch between knowledge of what is required for a given goaland the relevance of the subsequent behaviours to that goal. Datingback at least as far as Luria's (1966) ground-breaking studies of the ef-fects of frontal lobe damage, it is clear that ability to produce goal rele-vant behaviour is not determined simply by the objective complexityor difficulty of task execution demands. Rather than an inability to per-form a task effectively, there is loosening of the relationship between agiven goal and the knowledge required to achieve it such that the re-quired action(s) are simply not implemented. This phenomenon canalso be observed on tasks with multiple constraints in the normal pop-ulation (referred to as goal neglect; Duncan et al., 1996), and is particu-larly common in the performance of participants at the lower end of theg distribution. Our findings, like those of Duncan and colleagues, indi-cate that g confers an advantage at the level of goal conceptualisationrather than (or in addition to) task execution. Thus, it is not so muchthe real time processing demands of a task that drive performance–gcorrelations, but the efficiency with which the task-relevant constraintsare modelled prior to, and perhaps remodelled during, actualperformance.

An intuitive implication of our claim is that if the components (in-cluding response biases) inherent in a given task, however complex,are modelled in the same way by all participants, differences in perfor-mance would not be predicted in anymeaningful way by a participant'srelative position on the g distribution. Although it is impossible to deter-mine (and control how) a task is modelled with absolute certainty, weconsider this to be unlikely. In our view, variability in the mental con-struction/specification of complex goals is a major contributory factordistinguishing low from high g participants, but evidence for significantperformance–g correlations in basic tasks with minimal execution de-mands (e.g., simple reaction time and inspection time paradigms;Grudnik & Kranzler, 2001; Jensen, 1998, 2006; Kranzler & Jensen,1989; Sheppard & Vernon, 2008) suggests that factors associated withreal time task execution place demands on g over and above recruit-ment associated with efficiency of task modelling. Nevertheless, in ourstudy, as in that of Duncan et al. (2008), the body of instructions ratherthan the task per se appeared to drive sensitivity of performance to var-iations in g. In the current study, the status of task modelling in this re-lationship over more objective task characteristics is best illustrated bythe observation (in Experiment 4) that the response inhibition require-ment showed significantly closer sensitivity to g in the four-rule (rela-tive to two-rule) task, despite a near identical level of error in thetasks. In those trials incorporating response inhibition (i.e., double-matching frames), error rates were higher on most measures of perfor-mance (irrespective of instructions) and this was particularly true forindividuals who produced more errors on the Culture Fair. Althoughthis is consistent with evidence that inhibition is a primary risk factorfor recruitment of g, the correlational findings of Experiment 4 showthat the strength of this relationship is determined in large part by theefficiency of task conceptualisation.

A recent study (Bhandari & Duncan, 2014) has provided evidenceagainst the possibility that task instructions shape themental represen-tation of a novel task, as suggested by our findings. They found perfor-mance to be similar in a group given instructions chunked intodistinct parts and a group given instructions in a more “interleaved”fashion. Their investigations instead suggested that goal neglectwas de-termined by characteristics of the performed task, with the more com-plex sub-parts of the task (those with a greater number of taskcomponents) more likely associated with goal neglect, particularly in


people at the lower end of the g distribution. They interpreted thesefindings as evidence for goal-directed behaviour being realised via dis-crete “attentional episodes” characterised by a clear, focused associationbetween current demands of the environment and the appropriate be-havioural response, suggesting that fluid intelligence may reflect theability to convert complex requirements into “smaller more manage-able parts” (Bhandari & Duncan, 2014, p. 29; see also Duncan, 2013).From such a perspective we may have expected to see better perfor-mance and lower correlation with Culture Fair scores in our four-rulecondition, with the rules more distinctly mapping a specific task eventrelative to two-rule. It could be that the effects of instructions on taskmodelling are observed in tasks in which distinct task events are moreeasily mapped to distinct rules, such as in the feature match taskemployed here (as well as Duncan et al. 2008 and Roberts &Anderson, 2014; see also Dreisbach et al., 2007) but not in tasks inwhich distinct task parts involve a number of constraints. We arguethat the contribution of g to cognitive performance may reflect the effi-cacy with which an inefficient model can be reconceptualised towardsgreater efficiency,with efficiency being defined as fewer distinct chunks(however complex), thereby releasingWMC for attentional reallocation(rather than splitting the task intomanageable parts). Given the prelim-inary evidence for reconceptualisation of task information observed inour Experiment 4, we suggest that the possible link between g and theability to flexibly manipulate and remodel constraints associated withcomplex goal-directed behaviours is likely to be a valuable focus for fu-ture investigation.

Whatever the case, we submit that models of intelligence built oncorrelations with objective measures of task difficulty which do not in-corporate individual differences in the way in which those tasks arementally represented are overlooking a fundamental role of intelli-gence: to construct action plans and realise goals through optimal allo-cation of resources. Participants presented with the same complex taskmay not, from a cognitive perspective, be undertaking the same task.High g confers an ability to remove redundant task information and tochunk task relevant informationmore efficiently. The role during execu-tion may relate more to monitoring task content and manipulating themodel in the direction of greater accuracy and efficiency. Failure totake into account potential variability in the ways in which complextasks are conceptualised may thus limit the validity of theoretical infer-ences drawnon the basis of group averaging of performance scores. Fur-ther clarification of these and other possible explanatory variablesdriving individual differences in task performance will require the in-corporation of multiple indicators and the use of confirmatory factoranalysis or structural-equation modelling.

An area demanding attention in future work is the exact mecha-nisms by which fluid intelligence and task conceptualisation are con-nected, including explorations of potential alternative explanationsandmediating factors. This is particularly true in light of otherwork pro-viding evidence against the notion that memory and concurrent pro-cessing share a limited resource (e.g., Oberauer et al., 2004). Forexample, constructing and remodelling task representations could relyon processes akin to those in traditional conceptualisations of short-term memory. Our results do not support the notion that difficulties insuccessfully performing the experimental tasks were due to short-term memory failures in the sense of forgotten task instructions. It isnonetheless conceivable that the observed relationship between fluidintelligence and the way in which a task is cognitively modelled maydissolve when short-term storage capacity is controlled (in the sameway as the relationship between fluid intelligence and WM processingfactors such as attention control; e.g., Chuderski et al., 2012; Colomet al., 2008).

A recent theoretical development of the WM model distinguishesbetween declarative and procedural WM (Oberauer, 2009; Oberauer,Souza, Druey & Gade, 2013; see also Baddeley, 2012) with the latterserving the formation of goal-relevant stimulus-response bindings,and the bindings themselves forming the content of a higher level

executive system (the bridge; Oberauer, 2009). The model provides aframework for quickly establishing new stimulus-response associationsdefined by task instructions, and a ‘response focus’ necessary for settinga given response apart from other possible competing responses withinthe task set. Thus, the model incorporates inhibitory links between re-sponse alternatives within a task set and a mechanism for excludingtask irrelevant information (i.e., unless information is held in the bridge,candidate actions are unavailable for selection by the response focus).While thismodel requires further development, it provides an intuitive-ly appealing theoretical framework for advancing our understanding ofindividual differences in cognitive ability. In our view, exploration of therelationship between task performance and the capacity/functionalityof procedural WM is likely to foster important advances in this area.

In summary, the present findings indicate that the involvement ofSpearman's g in task performance is largely determined by the relation-al complexity of individual task components, and that the level of effortrequired for optimalmodelling of a task largely determines the strengthof performance–g correlations. Performance discrepancies betweenhigh and low g individuals are likely to be greatest in earlier stages ofnovel task execution when initial modelling of task constraints may besubject to reorganisation and consolidation, with the effectiveness andtime course of model stabilisation contingent upon the complexity ofthe task. Response conflict is an important risk factor for the engage-ment of g because it demands a clear parsing of available but incompat-ible responses in line with task instruction. Other constraints limit theability to form a stable task model, with those at the lower end of theg distribution less able to manipulate the representation of (and rela-tionship between) task demands in an optimal way. Future researchwill require systematic exploration of individual differences in theways in which task instructions are initially modelled and then re-modelled over the course of preparation for, and execution of, goal-directed complex behaviours.

Acknowledgements

Experiments 1–3 were supported by a Medical Research Council(MRC) PhD studentship awarded to PB while based at the MRC Cogni-tion & Brain Sciences Unit. Experiment 4was supported by an internallyfunded PhD studentship awarded to EC while based at Anglia RuskinUniversity. We wish to thank John Duncan for his encouragement andadvice during the course of this work, and an anonymous reviewer forhis/her helpful comments on an earlier draft.

References

Ackerman, P. L., Beier, M. E., & Boyle, M. O. (2002). Individual differences in workingmemory within a nomological network of cognitive and perceptual speed abilities.Journal of Experimental Psychology: General, 131(4), 567–589.

Baddeley, A. (2012). Working memory: theories, models, and controversies. AnnualReview of Psychology, 63, 1–29.

Baddeley, A. D., & Hitch, G. J. (1974). Working memory. In G. A. Bower (Ed.), Recent Ad-vances in Learning and Motivation (pp. 47–90). New York: Academic Press.

Bhandari, A., & Duncan, J. (2014). Goal neglect and knowledge chunking in the construc-tion of novel behaviour. Cognition, 130, 11–30.

Bright, P. D. (1998). The Control of Action: An Exploration of Spearman's General Factor. Un-published doctoral thesis Cambridge, England: University of Cambridge.

Broadbent, D. (1958). Perception and Communication. London: Pergamon Press.Cattell, R. B. (1971). Abilities: Their structure, growth, and action. Boston: Houghton‐Mifflin.Cattell, R. B., & Cattell, H. E. P. (1973). Measuring Intelligence With the Culture Fair Tests.

Champaign, IL: Institute for Personality and Ability Testing.Chuderski, A., Taraday, M., Nęcka, E., & Smoleń, T. (2012). Storage capacity explains fluid

intelligence but executive control does not. Intelligence, 40, 278–295.Cohen, J. D., MacWhinney, B., Flatt, M., & Provost, J. (1993). PsyScope: a new graphic in-

teractive environment for designing psychology experiments. Behavioral ResearchMethods, Instruments, and Computers, 25(2), 257–271.

Colom, R., Abad, F. J., Quiroga, M.Á., Shih, P. C., & Flores-Mendoza, C. (2008). Workingmemory and intelligence are highly related constructs, but why? Intelligence, 36(6),584–606.

Colom, R., Flores-Mendoza, C., Quiroga, M.Á., & Privado, J. (2005). Working memory andgeneral intelligence: the role of short-term storage. Personality and IndividualDifferences, 39(5), 1005–1014.

http://refhub.elsevier.com/S0001-6918(16)30126-3/rf0005



























Conway, A. R. A., Cowan, N., Bunting, M. F., Therriault, D. J., & Minkoff, S. R. B. (2002). Alatent variable analysis of working memory capacity, short-term memory capacity,processing speed, and general fluid intelligence. Intelligence, 30, 163–183.

Conway, A. R. A., Kane, M. J., & Engle, R. W. (2003). Working memory capacity and its re-lation to general intelligence. Trends in Cognitive Sciences, 7(12), 547–552.

Daneman, M., & Carpenter, P. A. (1980). Individual differences in working memory andreading. Journal of Verbal Learning and Verbal Behavior, 19(4), 450–466.

Das, J. P. (2002). A better look at intelligence. Current Directions in Psychological Science,11(1), 28–33.

Dempster, F. N., Corkill, A. J., & Jacobi, K. (1995, November). Individual differences in resis-tance to interference. Paper presented at the Annual Meeting of the Psychonomic Socie-ty, Los Angeles.

Desimone, R., & Duncan, J. (1995). Neural mechanisms of selective visual attention.Annual Review of Neuroscience, 18(1), 193–222.

Dreisbach, G., Goschke, T., & Haider, H. (2007). The role of task rules and stimulus–response mappings in the task switching paradigm. Psychological Research, 71(4),383–392.

Dumontheil, I., Thompson, R., & Duncan, J. (2010). Assembly and use of new task rules infronto-parietal cortex. Journal of Cognitive Neuroscience, 23(1), 168–182.

Duncan, J. (2013). The structure of cognition: attentional episodes in mind and brain.Neuron, 80, 35–50.

Duncan, J., Emslie, H., Williams, P., Johnson, R., & Freer, C. (1996). Intelligence and thefrontal lobe: the organization of goal-directed behavior. Cognitive Psychology, 30(3),257–303.

Duncan, J., Parr, A., Woolgar, A., Thompson, R., Bright, P., Cox, S., ... Nimmo-Smith, I.(2008). Goal neglect and Spearman's g: competing parts of a complex task. Journalof Experimental Psychology: General, 137(1), 131–148.

Duncan, J., Schramm, M., Thompson, R., & Dumontheil, I. (2012). Task rules, workingmemory, and fluid intelligence. Psychonomic Bulletin & Review, 19(5), 864–870.

Engle, R. W., & Kane, M. J. (2004). Executive attention, working memory capacity, and atwo-factor theory of cognitive control. In B. Ross (Ed.), The Psychology of Learningand Motivation (pp. 145–199). New York: Academic Press.

Engle, R. W., Tuholski, S. W., Laughlin, J. E., & Conway, A. R. (1999). Working memory,short-term memory, and general fluid intelligence: a latent-variable approach.Journal of Experimental Psychology: General, 128(3), 309–331.

Friedman, N. P., Miyake, A., Corely, R. P., Young, S. E., DeFries, J. C., & Hewitt, J. K. (2006).Not all executive functions are related to intelligence. Psychological Science, 17(2),172–179.

Fukuda, K., Vogel, E., Mayr, U., & Awh, E. (2010). Quantity, not quality: the relationship be-tween fluid intelligence and working memory capacity. Psychonomic Bulletin &Review, 17(5), 673–679.

Gray, J. R., Chabris, C. F., & Braver, T. S. (2003). Neural mechanisms of general fluid intel-ligence. Nature Neuroscience, 6(3), 316–322.

Grudnik, J. L., & Kranzler, J. H. (2001). Meta-analysis of the relationship between intelli-gence and inspection time. Intelligence, 29(6), 523–535.

Hasher, L. (1997). Inhibitory control over no-longer-relevant information: adult age dif-ferences. Memory & Cognition, 25(3), 286–295.

Hasher, L., & Zacks, R. T. (1988). Working memory, comprehension, and aging: a reviewand a new view. In G. H. Bower (Ed.), The Psychology of Learning and Motivation.Vol. 22. (pp. 193–225). San Diego, CA: Academic Press.

Hester, R., & Garavan, H. (2005). Working memory and executive function: the influenceof content and load on the control of attention.Memory & Cognition, 33(2), 221–233.

Jensen, A. R. (1998). The g Factor: The Science of Mental Ability. Westport, CT: Praeger.Jensen, A. R. (2006). Clocking the Mind: Mental Chronometry and Individual Differences. Ox-

ford: Elesvier.Kane, M. J., & Engle, R. W. (2003). Working-memory capacity and the control of attention:

the contributions of goal neglect, response competition, and task set to Stroop inter-ference. Journal of Experimental Psychology: General, 132(1), 47–70.

Kane, M. J., Conway, A. R. A., Hambrick, D. Z., & Engle, R. W. (2007). Variation in workingmemory as variation in executive attention and control. In A. R. A. Conway, C. Jarrold,M. J. Kane, A. Miyake, & J. N. Towse (Eds.), Variation in Working Memory (pp. 21–48).Oxford: Oxford University Press.

Kerns, J. G., Cohen, J. D., MacDonald, A. W., III, Cho, R. Y., Stenger, A., & Carter, C. S. (2004).Anterior cingulate conflict monitoring and adjustments in control. Science, 303,1023–1026.

Kievit, R. A., Davis, S. W., Mitchell, D. J., Taylor, J. R., Duncan, J., Cam-CAN, & Henson, R. N.(2014). Distinct aspects of frontal lobe structure mediate age-related differences influid intelligence and multitasking. Nature Communications, 5, 5658.

Kranzler, J. H., & Jensen, A. R. (1989). Inspection time and intelligence: a meta-analysis.Intelligence, 13(4), 329–347.

Long, D. L., & Prat, C. S. (2002). Working memory and Stroop interference: an individualdifferences investigation. Memory & Cognition, 30(2), 294–301.

Luria, A. R. (1966). Higher Cortical Functions in Man. London: Tavistock.Luria, A. (1973). The Working Brain. London: Penguin.Lustig, C., May, C. P., & Hasher, L. (2001). Working memory span and the role of proactive

interference. Journal of Experimental Psychology: General, 130(2), 199–207.May, C. P., Hasher, L., & Kane, M. J. (1999). The role of interference in memory span.

Memory & Cognition, 27(5), 759–767.Mitchell, J. P., Macrae, C. N., & Gilchrist, I. D. (2002). Working memory and the suppres-

sion of reflexive saccades. Journal of Cognitive Neuroscience, 14(1), 95–103.Miyake, A., & Friedman, N. P. (2012). The nature and organization of individual differ-

ences in executive functions four general conclusions. Current Directions inPsychological Science, 21(1), 8–14.

Miyake, A., Friedman, N. P., Emerson,M. J., Witzki, A. H., Howerter, A., &Wager, T. D. (2000).The unity and diversity of executive functions and their contributions to complex“frontal lobe” tasks: a latent variable analysis. Cognitive Psychology, 41(1), 49–100.

Morse, C. K. (1993). Does variability increase with age? An archival study of cognitivemeasures. Psychology & Aging, 8(2), 156–164.

Oberauer, K. (2009). Design for a workingmemory. Psychology of Learning andMotivation,51, 45–100.

Oberauer, K., Lange, E., & Engle, R. W. (2004). Workingmemory capacity and resistance tointerference. Journal of Memory and Language, 51(1), 80–96.

Oberauer, K., Souza, A. S., Druey, M. D., & Gade, M. (2013). Analogous mechanisms of se-lection and updating in declarative and procedural working memory: experimentsand a computational model. Cognitive Psychology, 66(2), 157–211.

Polderman, T. J., de Geus, E. J., Hoekstra, R. A., Bartels, M., van Leeuwen, M., Verhulst, F. C.,... Boomsma, D. I. (2009). Attention problems, inhibitory control, and intelligenceindex overlapping genetic factors: a study in 9-, 12-, and 18-year-old twins.Neuropsychology, 23(3), 381–391.

Rabbitt, P. (1993). Does it all go together when it goes? The Nineteenth Bartlett MemorialLecture. Quarterly Journal of Experimental Psychology, 46(A)(3), 385–434.

Redick, T. S., Calvo, A., Gay, C. E., & Engle, R. W. (2011). Working memory capacity and go/no-go task performance: selective effects of updating, maintenance, and inhibition.Journal of Experimental Psychology: Learning, Memory, and Cognition, 37(2), 308–324.

Roberts, G., & Anderson, M. (2014). Task structure complexity and goal neglect in typical-ly developing children. Journal of Experimental Child Psychology, 120, 59–72.

Roberts, R. J., Jr., & Pennington, B. F. (1996). An interactive framework for examining pre-frontal cognitive processes. Developmental Neuropsychology, 12(1), 105–126.

Roberts, R. J., Hager, L. D., & Heron, C. (1994). Prefrontal cognitive processes: workingmemory and inhibition in the antisaccade task. Journal of Experimental Psychology:General, 123(4), 374–393.

Salthouse, T. A. (1992). Reasoning and spatial abilities. In F. I. M. Craik, & T. A. Salthouse(Eds.), The Handbook of Aging and Cognition. New Jersey: Lawrence Erlbaum.

Salthouse, T. A., Atkinson, T. M., & Berish, D. E. (2003). Executive functioning as a potentialmediator of age-related cognitive decline in normal adults. Journal of ExperimentalPsychology: General, 132(4), 566–594.

Sheppard, L. D., & Vernon, P. A. (2008). Intelligence and speed of information-processing:a review of 50 years of research. Personality and Individual Differences, 44(3),535–551.

Süß, H., Oberauer, K., Wittmann, W. W., Wilhelm, O., & Schulze, R. (2002). Working-memory capacity explains reasoning ability—and a little bit more. Intelligence,30(3), 261–288.

Turner, M. L., & Engle, R. W. (1989). Is working memory capacity task dependent? Journalof Memory and Language, 28(2), 127–154.

Unsworth, N., & Engle, R. W. (2007a). The nature of individual differences in workingmemory capacity: active maintenance in primary memory and controlled searchfrom secondary memory. Psychological Review, 114(1), 104–132.

Unsworth, N., & Engle, R. W. (2007b). On the division of short-term and working memo-ry: an examination of simple and complex span and their relation to higher orderabilities. Psychological Bulletin, 133(6), 1038–1066.

Unsworth, N., & Spillers, G. J. (2010). Working memory capacity: Attention control, sec-ondary memory, or both? A direct test of the dual-component model. Journal ofMemory and Language, 62(4), 392–406.

Unsworth, N., Redick, T. S., Lakey, C. E., & Young, D. L. (2010). Lapses in sustained attentionand their relation to executive control and fluid abilities: an individual differences in-vestigation. Intelligence, 38(1), 111–122.



































































































































involvement of spearman's g in conceptualisation versus...

Documents